Development of a parallel computing-based Futureland model for multiple land-use simulation: a case study in Shanghai

Figures & data

Figure 1. The neighborhood shape definition tool, and a case of defining the neighborhood weights and effects. For example, for land-use transition to type B, the weights for types A, B, C and D are w₁, w₂, w₃, and w₄, respectively. In a 3 × 3 square neighborhood, the total effects for type B are (3w₁+3w₂+w₃+2w₄)/8.

Figure 2. The definition of the transformation cost between any two land-use types.

Figure 3. The processing of edge enhancement in the Futureland model.

Table 1. The definition of assessment metrics of CA models and results.

Type	Metric	Definition
Simulation accuracy	Producer’s accuracy	Percentage of correctly simulated pixels in the ground reference results corresponding to a particular land-use type.
	User’s accuracy	The number of correctly simulated pixels associated with a particular land-use type, as a percentage of the predicted results for that type.
	Overall accuracy	The sum of correctly simulated pixels divided by the total number of pixels.
	Kappa simulation	The coefficient of agreement between the simulated land-use transitions and the actual land-use transitions (Van Vliet et al. 2011).
	Fuzzy Kappa	A variant of the Kappa coefficient that considers different degrees of similarity between pairs of cells on two maps (Hagen 2003).
	Aggregated cells	Similarities of simulated and actual land-use patterns at different scales (Pontius et al. 2004).

Figure 4. The study area of Shanghai: (a) Ecological spatial plan (2017–2035); (b) Land-use pattern in 2010.

Table 2. Land-use classification criteria and definitions in the Shanghai study area.

New category	Initial category	Definition
Urban land	Built-up land	Land for urban and rural settlement, and man-made buildings and projects
Water area	Water body	Land covered by natural water bodies such as rivers and lakes or land with water storage and irrigation functions
Greenspace	Woodland & grassland	Land covered with vegetation, including natural forests, grasslands, wetlands and artificially planted orchards
Farmland	Cropland	Farmland for crops, including paddy land and dry land
Special land	Other built-up land	Land such as airports and harbors

Table 3 summarizes the remote sensing images and geographic information data used in this study. In addition to Landsat images, we collected Digital Elevation Model (DEM; from ASTER GDEM 2) data from the Geospatial Data Clouds. The socioeconomic data include the GDP density provided by CAS and the population density provided by Worldpop. The POI datasets including road and river networks were collected from OpenStreetMap. We performed a uniform coordinate transformation of all datasets using ArcGIS 10.2, then extracted 11 factors from the above dataset by following earlier works (Cloern et al. 2016; Chen et al. 2017; Gao et al. 2020). All distance factors were extracted from the facility data, such as various POIs, roads and rivers. All factors were converted to grid data with a resolution of 30 meters by resampling.

Table 3. Remote sensing images and geographical information data used in this study.

Type	Dataset	Factor	Scale	Date	Data source
Land	Landsat-5 TM	N/A	30m	2010-12-27	earthexplorer.usgs.gov
	Landsat-8 OLI	N/A	30m	2015-03-12
	Landsat-8 OLI	N/A	30m	2020-02-22
Terrain	ASTER GDEM 2	DEM	30m	2015	www.gscloud.cn
Socioeconomic data	Population density	POP	1000m	2015	www.worldpop.org
	Gross domestic product	GDP	1000m	2015	www.resdc.cn
Facilities	City center POI	D-city	30m	2020	www.openstreetmap.org
	District POI	D-district
	Town center POI	D-town
Roads and rivers	Primary road	D-road
	Highway	D-highway
	Subway	D-subway
	Railway	D-railway
	River	D-river

Figure 5. Futureland for complex land-use change simulation and future scenario projections.

Figure 6. The probability-of-occurrence maps for the five land-use types in Shanghai.

Figure 7. The simulated land-use patterns for 2015 and 2020 in comparison to the actual patterns.

Figure 8. Future land-use scenarios of Shanghai in 2025 to 2035 projected using Futureland.

Table 4. The coefficients of the factors tested for statistical significance.

Factor	Urban land	Water area	Greenspace	Farmland	Special land
D-city	2.08	−2.18	/	3.04	/
D-district	−0.87	2.57	/	/	/
D-town	−1.15	/	/	/	/
D-road	/	/	3.87	/	/
D-highway	−0.73	/	−1.21	1.95	/
D-subway	−1.51	/	/	/	/
D-railway	2.22	/	−0.66	/	/
D-river	−2.26	/	/	−4.19	/
DEM	1.96	−7.20	−4.95	6.06	/
POP	−3.91	/	−12.00	/	/
GDP	−4.21	−117.75	−39.76	/	/

Inputting the land-use maps (both 2010 and 2015) and the most relevant factors into Futureland, we generated the probability-of-occurrence of the five land-use types (Figure 6). The probability maps reflect the possibility of each cell transforming into the relevant land-use type. The probability of urban land (urban sprawl) is relatively concentrated, as most of the high values are outside the city center and around the Outer Ring Highway. The low average and maximum probabilities for the waters are probably because they have not changed much in the last decade. The high probability of greenspace is mainly concentrated in the northeast coastal and southwest areas. Although the average probability of farmland is also low, the high probability areas are mainly distributed outside the Suburban Ring Expressway. The probability of special land is zero throughout the study area because of their weak variation.

Table 5. The definition of the neighborhood weight matrix.

The central cell type	The neighboring cell type
	Urban land	Water area	Greenspace	Farmland	Special land
Urban land	1.0	0.1	0.2	0.5	0.1
Water area	0.1	1.0	0.1	0.1	0.1
Greenspace	0.1	0.2	1.0	0.2	0.1
Farmland	0.1	0.1	0.1	1.0	0.1
Special land	0.1	0.1	0.1	0.1	1.0

Table 6. The definition of the transformation cost matrix.

Transformation from	Transformation into
	Urban land	Water area	Greenspace	Farmland	Special land
Urban land	0.0	1.0	1.0	1.0	1.0
Water area	1.0	0.0	0.2	0.9	1.0
Greenspace	0.2	0.9	0.0	0.9	1.0
Farmland	0.1	0.8	0.2	0.0	1.0
Special land	1.0	1.0	1.0	1.0	0.0

In this study, we used the 2010 land-use pattern (Figure 7a) as the start map and the 2015 pattern (Figure 7c) as the end map to calibrate the Futureland model. Visual inspection reveals that the simulated and actual land-use patterns are overall similar, with rapid urban sprawl from the city center to the suburbs over the past decade. We selected an area north of the city center and along the Huangpu River and zoomed on them to show more details. Figure 7a1–e1 display that farmland was obviously occupied by urban land while urban greenspace was gradually decreasing. This indicates that Futureland has effectively captured the trend of urban sprawl, but there are still some mis-simulations, such as the greenspace in the upper right corner (c.f. Figure 7c1).

Table 7. The quantitative assessment of the simulated results for both calibration and validation.

Peroid	Type	Producer’s accuracy (%)	User’s accuracy (%)	Overall accuracy (%)	Kappa simulation	Fuzzy Kappa	Aggregated cells
2010–2015 (Calibration)	Urban land	85.0	82.6	0.866	0.796	0.801	0.894
	Water area	89.1	95.9
	Greenspace	77.1	77.0
	Farmland	89.1	89.0
	Special land	100.0	100.0
2015–2020 (Validation)	Urban land	85.8	84.3	0.845	0.780	0.783	0.886
	Water area	92.1	94.1
	Greenspace	68.2	72.2
	Farmland	84.1	84.0
	Special land	100.0	100.0

Table 8. Total land demand in 2025, 2030 and 2035 based on modified Markov chain for Shanghai.

Year/Period	Land-use type (%)
	Urban land	Water area	Greenspace	Farmland	Special land
2020	40.3	12.4	8.3	38.5	0.5
2025	43.9	12.2	7.3	36.1	0.5
2030	47.3	12.0	6.4	33.8	0.5
2035	50.5	11.6	5.8	31.6	0.5
2020–2025 Change	3.6	−0.2	−1.0	−2.4	0.0
2020–2030 Change	7.0	−0.4	−1.9	−4.7	0.0
2020–2035 Change	10.2	−0.8	−2.5	−6.9	0.0

Figure 8 shows the future land-use scenarios projected using the validated Futureland model and the spatial dynamics of future land-use change, showing primary land-use change occurring in the southern part. The projection shows that urban sprawl in Shanghai will likely occur around the urban fringes and mainly occupies farmland; however, this does not make much change in the overall land-use patterns over the next 15 years. Meanwhile, urban sprawl outside Shanghai’s Outer Ring Highway and around the Suburban Ring Highway will be more significant than in other areas of Shanghai.

Two typical areas were chosen to show the details of the land-use scenario over the next 15 years (Figure 8a1–c2), where the first is around Wujing Town in the southern part (enlarged area-1) and the second is in the southeastern-most part of the Dishui Lake area (enlarged area-2). In Wujing Town, the northwestern area along the Huangpu River is close to the edges of the existing built-up area, where the land type changes mainly from farmland to urban land. In Dishui Lake, more greenspace is converted to urban type, lying around the Dishui Lake, while farmland in the north and west will be gradually occupied and some greenspace near the sea will be converted to urban type.

Figure 9. Buffer-based analysis of urban sprawl on farmland and greenspace in 2025, 2030, and 2035.

Figure 10. Urban land and greenspace scenarios for 2035. (a) Urban land scenario associated with the four eco-space types in urban planning; (b) greenspace scenario associated with the four eco-space types; (c) the proportion of land-use types associated with the four eco-space types.

Van Vliet J, Bregt AK, Hagen-Zanker A. 2011. Revisiting Kappa to account for change in the accuracy assessment of land-use change models. Ecol Modell. 222(8):1367–1375.

 Web of Science ®Google Scholar

Hagen A. 2003. Fuzzy set approach to assessing similarity of categorical maps. Int J Geog Inf Sci. 17(3):235–249.

 Web of Science ®Google Scholar

Pontius RG Jr, Huffaker D, Denman K. 2004. Useful techniques of validation for spatially explicit land-change models. Ecol Modell. 179(4):445–461.

 Web of Science ®Google Scholar

Cloern JE, Abreu PC, Carstensen J, Chauvaud L, Elmgren R, Grall J, Greening H, Johansson JOR, Kahru M, Sherwood ET, et al. 2016. Human activities and climate variability drive fast‐paced change across the world’s estuarine–coastal ecosystems. Glob Chang Biol. 22(2):513–529.

 PubMed Web of Science ®Google Scholar

Chen Y, Liu X, Li X. 2017. Calibrating a land parcel cellular automaton (LP-CA) for urban growth simulation based on ensemble learning. Int J Geog Inf Sci. 31(12):2480–2504.

 Web of Science ®Google Scholar

Gao C, Feng Y, Tong X, Lei Z, Chen S, Zhai S. 2020. Modeling urban growth using spatially heterogeneous cellular automata models: comparison of spatial lag, spatial error and GWR. Comput Environ Urban Syst. 81:101459.

 Web of Science ®Google Scholar

Data availability statement

The Futureland software we developed is available at www.researchgate.net/publication/356978377 and the data we used is available at figshare.com/s/af1c7c14170fc44c39eb.